Recently I have been reading Ruth Byrne’s book The Rational Imagination (2005). The book turned out less relevant to the things I am interested in. To make sure it wasn’t a total waste though, I would like to raise a worry I have with the general argumentative strategy of the book.
The Rational Imagination is really two books in one. The descriptive book summarizes many interesting results of the psychology experiments Byrne and her associates have done on how people’s counterfactual reasoning tends to be influenced. When thinking about how things might be different, people tend to focus on short-term consequences of actions, long-term consequences of inactions, controllable events, and enabling (as opposed to causal) relations. Ch. 3-7 presents interesting empirical results that should be of interest to philosophers interested in modal epistemology. The normative book promises to argue that counterfactual reasoning is rational, but I am not sure she delivers on this promise. I will raise a worry for her argument, and from that, suggest some things she would need to explain in order to spell out a more complete theory of rationality for counterfactual reasoning.
The stated overarching argument of the book is as follows (208):
1. Humans are capable of rational thought.
2. The principles that underlie rational thought guide the sorts of possibilities that people think about.
3. These principles underlie counterfactual imagination.
C. Counterfactual imagination is rational.
(I prefer the term ‘counterfactual reasoning’ rather than her ‘counterfactual imagination’ since her experiments test how people make inferences from counterfactuals as opposed to factual (indicative) conditionals. Her use of the term, I suspect, is to differentiate the imagination in counterfactual reasoning from other kinds, such as in imagery or in our engagement with fiction.)
Back to her argument, I am uncertain of its soundness. It is unclear what ‘underlie’ means. It could be the case that we are able to make certain inferences due to some mental process, but that mental process does not explain why those inferences are rational to make. In which case, having the same underlying mental processes would not guarantee also being rational. We need either a theory of the internal architecture of how “rational thought” and counterfactual reasoning respond to reasons, or some external criteria that show the processes do equally well, where well means something like in accordance with our epistemic norms.
Byrne offers neither. In fact, her discussion of rationality in the conclusion is surprisingly tentative. It seems very strange that you could say that counterfactual reasoning is rational, without specifying what it means to be rational. In fact, her experimental results suggest that it fails to satisfy some intuitive external criteria. (She seems to take reasoning with factual conditionals as exemplars of uncontested “rational thought”. I am not sure about this move, but let us set that worry aside.)
Here is an experiment she describes (49-50). Half of the participants were given the factual conditional ‘if Linda is in Galway then Cathy is in Dublin.’ The other half were given the counterfactual ‘if Linda were in Galway then Cathy would be in Dublin.’ Then they are given premises ‘Linda is in Galway’, ‘Cathy is in Dublin’, ‘Linda is not in Galway’, ‘Cathy is not in Dublin’, and asked what, if any, followed from the premises. The results indicate people are about equally good at making modus ponens from both conditionals, but they make twice as many modus tollens inferences with the counterfactual than with the factual conditional. The explanation is that people tend to keep two possibilities in mind with counterfactuals (A + B and not-A + not-B) whereas they only keep one possibility in mind with factual conditionals (A + B). So they make the inferences from not-B to not-A more readily, since that possibility is already mentally represented. That’s the good news.
But if the two-possibilities explanation is right, then it should also predict that people will make denial of antecedent inferences from not-A to not-B more readily. The experiment result confirms this; in fact, people also make twice as many denial of antecedent inferences from counterfactuals than from factual conditionals. But this is bad, if we think denial of antecedent is an epistemically irrational inference procedure. For the record, people also make slightly more affirmation of consequent inferences with counterfactuals, which also looks bad. (I should note that the same worry is independently raised in Anand Jayprakash Vaidya’s review of the book in Philosophical Psychology 20:2 (2007).)
Then her explanation begins to look implausible from an evolutionary point of view. If counterfactual reasoning is less reliable, or even as reliable, as factual reasoning, but uses more mental resources by representing more possibilities, then it seems unlikely we would develop this ability. It is better to conserve the mental resources and only have the factual reasoning process. Sure, we would miss a lot of modus tollens, but we would also avoid many denial of antecedent inferences. However, it is clear that counterfactual reasoning plays an important role in our mental lives, e.g. planning, so there must something wrong with her two-possibilities explanation. So the objection goes.
There are two compatible ways that she might answer this objection. I am indebted to Aaron Bronfman for discussion here. First, she could show that the counterfactual reasoning process postulated by her theory nevertheless has high ecological validity. That is, in real-life situations there are more instances where we would use modus tollens than denial of antecedent. This is of course an empirical matter. Secondly, she could explain why denial of antecedent is not an epistemically bad inference. Doing that, of course, demands her to say what is the notion of rationality that she has in mind. It makes sense to spell out the criteria of rationality first, then test whether the counterfactual inferences people make are in fact rational.
Here is one sketchy proposal for why denial of antecedent is not bad, at least in some cases. It seems that it is okay in some circumstances to infer from ‘if Linda were in Galway then Cathy would be in Dublin’ and ‘Linda is not in Galway’ to ‘Cathy is in Dublin’. What appeared to be a denial of antecedent might really be a valid inference from some pragmatic content, presupposition, or background belief. It might be that whenever people talk about others’ whereabouts, they mean to imply a causal connection both ways. So from ‘if Linda were in Galway then Cathy would be in Dublin’ we infer ‘if Cathy were in Dublin then Linda would be in Galway’, and then we perform a modus ponens inference. Even supposing that this type explanation could be made to work in all cases, figuring out the details would still be quite difficult. So I will leave that task to someone else.